zuazo/whisper-medium-pt

Whisper Medium Portuguese

This model is a fine-tuned version of openai/whisper-medium on the mozilla-foundation/common_voice_13_0 pt dataset. It achieves the following results on the evaluation set:

• Loss: 0.2733
• Wer: 7.1156

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 1e-05
• train_batch_size: 64
• eval_batch_size: 32
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 20000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.0335	3.52	1000	0.2096	7.4771
0.0073	7.04	2000	0.2733	7.1156
0.0049	10.56	3000	0.2880	7.2520
0.0031	14.08	4000	0.3146	7.6233
0.0025	17.61	5000	0.3191	7.7416
0.0012	21.13	6000	0.3269	7.7942
0.0022	24.65	7000	0.3430	7.9141
0.0018	28.17	8000	0.3581	8.0242
0.0007	31.69	9000	0.3701	8.1556
0.001	35.21	10000	0.3582	8.0390
0.0002	38.73	11000	0.3851	7.8467
0.0004	42.25	12000	0.3890	8.1622
0.0001	45.77	13000	0.3757	8.0636
0.0	49.3	14000	0.4009	7.8895
0.0	52.82	15000	0.4136	7.8352
0.0	56.34	16000	0.4231	7.8106
0.0	59.86	17000	0.4311	7.7432
0.0	63.38	18000	0.4380	7.7022
0.0	66.9	19000	0.4430	7.6808
0.0	70.42	20000	0.4452	7.6644

Framework versions

• Transformers 4.37.2
• Pytorch 2.0.1+cu117
• Datasets 2.14.4
• Tokenizers 0.15.1